Ten years ago, I got Snowy on Craigslist.

Snowy is a toy poodle about the size of a good dictionary, with the energy of something much larger and the eyes of someone who had already figured out how to get what she wanted. Within a week, she had learned to sit, to stay, and to fetch. Within a month, she had learned that fetching got her attention, and that attention was really what she was after.

I thought she was extraordinary. I showed her off constantly to friends, family and quite frankly to anyone who glanced at us walking in the neighborhood..

What I didn’t realize until much later was how much of our relationship I had built around giving her commands. Fetch. Stay. Come. She was remarkable at all of them(most of them, she does not always fetch). But I was the one who always decided what to fetch, where to go, and why.

Snowy just executed.

I’ve been thinking about Snowy a lot lately. Because I think most of us are in exactly the same relationship with AI.

Most of us are treating AI like a very impressive dog. We’re genuinely amazed by it and frankly it is amazing, and that we are the only ones with a great pet.. We teach it tricks we learned from X and LinkedIn. We show it off to friends and ask it to fetch things. Fetch me a summary. Fetch me an email to my boss. Fetch me five subject line options. And it fetches. Every time. Without chewing the furniture.

But we’re still the ones deciding what to fetch and the ones who know where we’re going and why. The dog is just faster than going by ourselves.

We have seen this movie before. It’s what happens with every technology that matters, and most people are using AI the way they used the Internet in 1995. They’re impressed by it and playing with it. They’re using it to do things they already did, slightly faster. But it’s worth understanding why, because the people who figure it out early tend to do very well.

When the Internet arrived, smart people immediately built online versions of offline things. They built magazines, stores and classified ads. Even the word “classified” was borrowed whole from print.

Before the Internet, they paid $4 a line in the Sunday paper and maybe twelve people in the zip code saw it. With the Internet, they posted the same three sentences to a website and maybe twelve thousand people saw it.

It was the same dog, and the same ad but a bigger audience.

That felt like a revolution. With that revolution, we saw progress that was real and measurable but it wasn’t insightful. The insight came later, when someone realized the Internet didn’t just distribute the ad faster but made the ad itself obsolete.

Google wasn’t an online version of the Yellow Pages. Amazon wasn’t Borders with a website. They were things that could only exist because the Internet existed. The distinction sounds simple but it took most people a decade to actually see it.

The iPhone repeated this almost exactly. For the first two years, people used it to check email and browse websites much the same things they did on computers, but from their couch. Then someone realized you had a GPS, a camera, a microphone, and a network connection in the pocket of every person on earth, and that this combination made entirely new things possible. Like Uber, Instagram and WhatsApp. None of them were faster versions of something old. They were new species.

AI is at the Craigslist moment right now.

What people are actually building

The tell is what’s getting made in the world today. And here is a sampling.

There are currently more than a dozen apps using AI to generate your daily horoscope. Aistro. Astroficial. VedicAstroGPT. ZodiacGPT and each is confidently telling you Mercury is in retrograde and they are all powered by the same technology that Geoffrey Hinton says will cause massive unemployment and reshape civilization.

SocialAI is an app where every account except yours is a bot. You post something, and your AI followers respond on cue: fans, skeptics, haters, optimists, whatever you select. A social network where you are the only human, surrounded by algorithms programmed to validate you. Someone built this, presumably pitched it to investors, and those investors wrote a check.

MeowTalk translates your cat’s meows. Not metaphorically. It listens and tells you what the cat means. It has hundreds of thousands of downloads and to be fair there are competing cat translator apps. Multiple teams looked at the state of AI and decided the most pressing unsolved problem was interspecies communication with house pets.

Sora is the most expensive version of this story. OpenAI launched a text-to-video app in September 2025 that hit number one on the App Store within 24 hours. Within weeks, people were using it to generate Mario smoking weed, Pikachu ASMR, and deepfakes of Martin Luther King Jr. Someone made Ronald McDonald do things Ronald McDonald should not do. The fetch commands were spectacular. Downloads peaked at 3.3 million in November 2025 then dropped to 1.1 million by February 2026, generating only $2.1 million in revenue against costs of roughly a million dollars a day.

OpenAI shut it down six months after launch — a money pit that nobody was using, burning through compute that could have been building something that mattered. Disney had committed a billion dollars to the partnership. Disney executives were informed of the closure less than an hour before the public announcement. The fetch was impressive but nobody asked what they were actually building.

Is this surprising? No….it’s what happens when a new capability meets human nature. The first thing people did with the printing press was print indulgences. The first things people did with the internet were download music illegally and look at things they couldn’t look at in public.

The problem is most people stop there. They keep sending fetch commands. Not because they’re lazy and Snowy was never lazy either! It is because the command worked, and the thing came back, and that felt like enough.

The problem isn’t that they’re fetching. The problem is that they’ve stopped asking whether they should be fetching at all and whether the thing they’re sending it after is actually the thing they need.

A dog that can fetch anything is extraordinary but it doesn’t know that you’re not thirsty. It doesn’t know the cup is already full. It doesn’t know you’ve been fetching the wrong thing for three years and the real problem is in the other room.

What the people who built this actually think

The researchers who created the technology are, at this point, fairly openly frustrated with how it’s being used.

Hinton spent fifty years working on neural networks. He left Google in 2023 specifically to be able to say what he actually thought. What he thinks is that AI will replace enormous numbers of jobs - not someday, but now, faster than most people realize. Every seven months or so, AI can do in half the time what it could do before. He’s not excited about this. But he’s clear: this is what’s happening whether people pay attention or not.

Yann LeCun, who ran AI research at Meta for years, makes a different but related point. Current AI, he argues, isn’t actually understanding anything - in fact it is predicting the next word, very well. But there’s a gap between predicting the next word and reasoning about the world, and most people don’t appreciate how large that gap is. When we figure out how to close it, the jump will be significant.

Terence Tao, probably the most talented mathematician alive, is more optimistic. His view is that AI is going to be a remarkable collaborator for people doing genuinely hard things. The key word is collaborator and the human still has to know what question to ask. The human still has to have the taste to recognize a good answer. What AI changes is the cost of executing once you know what you’re trying to do.

Three people who best understand what’s happening. Three different angles. Same basic point: this is much more powerful than most people are treating it, and they’re mostly using it wrong.

The actual opportunity

Here’s the part most people skip past because it’s uncomfortable.

If AI is compressing jobs and it is, already, then the default response is exactly wrong. The default response is to optimize for the old system. Write a better resume and get better at interviews, find the right keywords and polish LinkedIn. This makes sense if the system is stable but if the system itself is changing, optimizing for it is the worst possible strategy. You’re getting better at a game whose rules are being rewritten while you practice.

The alternative isn’t to panic but notice something the job-loss framing misses entirely.

AI didn’t just eliminate certain kinds of work. It eliminated the price of entry for building things. For most of history, having deep domain expertise and having the ability to build something around that expertise were two different things requiring two different people and usually a third person with money. That gap is closing fast.

The person who spent ten years in healthcare operations and knows exactly where the workflow breaks can now build the tool that fixes it. The marketing leader who understands a niche industry better than any outsider can now build the platform that industry needs. The supply chain expert who has watched the same inefficiency persist for a decade can now do something about it all alone, quickly, and without a team or a fundraise or permission from anyone.

This is not a consolation prize for people who lost jobs but a genuine structural shift in who gets to build things and solve hard problems. For most of human history, that was a very short list. It’s getting longer fast.

The people pulling ahead right now aren’t the ones with the best prompts but the ones who stopped giving fetch commands long enough to ask a different question: what would I do if I had something that could actually think alongside me - not just run fast in the direction I point?

Thinking alongside AI looks like bringing it into the room before you know the answer and not after. It looks like using it to pressure-test the strategy, not just produce the deliverable. It looks like saying: here’s what I’m trying to build, here’s what I know, here’s where I keep getting stuck and then actually listening to what comes back.

Most people use AI at the end of their thinking. They’ve already decided. They just need the words. That’s fetching.

The shift is using it at the beginning. Before the deck, email or plan. When the question is still open and the thinking is still soft.

The pattern

We didn’t get the full value of the Internet from portals. We got it from people who took the underlying capability seriously and asked what it made newly possible.

We didn’t get the full value of the iPhone from the email on the couch. We got it from people who looked at the hardware and realized it enabled entirely new things.

AI is not primarily a better way to write emails or generate horoscopes or build social networks for bots. It’s a significant reduction in the cost of building things and solving hard problems. The people who will look back on this moment as a turning point will be the ones who understood that early.

Snowy is eleven now. She still fetches and mostly good at it. But the best moments we’ve had together were never the fetch.They were the times she just sat next to me while I was figuring something out, and somehow that was enough.

The pet phase always ends. Here’s how to leave it.

1. Write down the one thing you’ve been putting off. This should not be a vague goal but a specific problem you actually know how to solve because of your years in the industry, your frustrations, your understanding of where the process breaks. That thing.

2. Open a conversation. Don’t ask it to fetch anything. Please don’t write “write me a summary.” Please stop asking it to “give me five ideas.” Start with: here’s what I’m trying to build, here’s what I know, here’s where I keep getting stuck.

3. Read what comes back like it’s a thought partner, not a search result. You’re not looking for the answer but for the thing you realize while reading it.

4. Start today. Specifically today. Not when you have more time and not after the next meeting.

The window is open and the people who look back on this as a turning point didn’t do something dramatic. They just started on a Tuesday afternoon when everyone else sent another fetch command.